Sustainable machining demands efficient tool condition monitoring (TCM) to maximize tool utilization and reduce environmental impact. Existing TCM solutions range from high-cost multi-sensor systems to ultra-low-cost alternatives with limited accuracy. This research bridges the gap with a resource-efficient, standalone TCM system for on-site tool wear estimation. The system integrates a PVDF-based accelerometer, Raspberry Pi 4, and a data acquisition card. A multi-level software architecture is designed to fully leverage this hardware, optimizing real-time signal processing while supporting both machine learning model training and inference. The proposed method employs two elementary models: k-means clustering for machining phase segmentation and ridge regression for tool wear estimation. A case study on an industrial lathe established a linear correlation between tool wear and surface roughness. Data from three tool inserts over their lifetimes proved sufficient for training machine learning models, achieving promising prediction accuracy. This research advances standalone TCM solutions tailored for manufacturing sectors seeking a balance between affordability and performance.